December 9th, 2011 To: Mr Abdul Waheed Haris Lecturer – Technical Report Writing FAST-NUCES (Main Campus) Karachi

Dear Sir, I have enclosed the Final Year Project Report of IEEE 802.15.4/ZigBee Network Lifetime Optimization. As I had predicted, the introduction of Cluster Heads or ZigBee Coordinators to the simulated network on OMNeT++ and adding a little more intelligence to them increased the network uptime by at least 10%. The details are discussed in the enclosed report. If the methodologies that I have discussed are implemented, it would be a great achievement for the WSN Committee to have an even better and efficient network at their disposal. It was a great honour working with you on this project. I would be available for contact on my email address if there are any queries regarding this project and/or project report.

Thank You.

Yours Sincerely,

Usman Muhammad Nooruddin Roll Number: 07-0039 Section A, BS (TE) Email Address:


National University of Computer and Emerging Sciences – FAST

Dated: December 9th, 2011

Prepared By: Usman Muhammad Nooruddin Roll Number: 07-0039

Prepared For: Abdul Waheed Haris Lecturer – Technical Report Writing

[This page is intentionally left blank]


Table of Contents
01. 02. 03. 04. 05. 06. 07. 08. 09. 10. Abstract Summary Introduction Background Scope Applicable Documents Nomenclature / Definitions References Methodology Theory 10.1 Evaluation of Low Rate Wireless Personal Area Network (LR-WPAN) Standardization 10.2 Why is it called ZigBee? 10.3 Device Types 10.4 Network Topologies 10.5 Architecture 10.6 Network and Application Support Layer 10.7 Physical (PHY) Layer 10.8 Media Access Control (MAC) Layer Design Parameters Technical Requirements Assumptions Technical Description Instructions & Procedures Evaluations / Analysis Acceptance Criteria Results & Conclusion Discussion of Results Recommendations & Alternatives Bibliography Appendix “Annealing Sensor Networks”, Andrew Jennings and Daud A. Channa List of Figures: Figure 1: Star Topology Figure 2: Peer – to – Peer Topology Figure 3: Mesh Topology Figure 4: ZigBee Stack Figure 5: Proposed Scenario List of Graphs: Graph 1: Results 15 9 9 10 10 12 4 5 5 5 6 6 7 7 7 8 8 8 8 9 10 11 11 11 11 12 12 13 13 14 14 15 15 15 16 17 18

11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.


IEEE 802.15.4 or ZigBee is a standards-based technology for remote monitoring, control and sensor network applications. The ZigBee standard was created to address the need for a cost-effective, standardsbased wireless networking solution that supports low data-rates, low-power consumption, security, and reliability. Although the ZigBee Alliance employs enough intelligence for a network to grant ZigBee devices a long uptime, this report studies another method that can well increase the Network Lifetime up to 10% - 15% more than the conventional methods that are proposed by default. This is done by intelligently routing the packets from source to sink by looking at the LQI (Link Quality Indicator) of the link currently under consideration.


This report focuses on the possibility of an introduction of a new protocol to route the packets from the ZigBee end-devices to the Monitoring station at the other end. A packet may have to pass many devices in between to reach the predefined destination. But, it all comes down to the cumulative intelligence of the network to route that packet efficiently from the source to the destination to save energy in this process.

The technology defined by the ZigBee specification is intended to be simpler and less expensive than other WPANs, such as Bluetooth. ZigBee is targeted at radio-frequency (RF) applications that require a low data rate, long battery life, and secure networking. This research report studies the implementation of a new Networking protocol that provides further intelligence to the ZigBee Coordinators so that they can well manage the routing of packets from source to sink. This report complies with the standards set by the IEEE 802.15.4.

The IEEE 802.15.4 standard is a simple packet data protocol for lightweight wireless networks and specifies the Physical (PHY) and Medium Access Control (MAC) layers for Multiple Radio Frequency (RF) bands, including 868 MHz, 915 MHz, and 2.4 GHz. The IEEE 802.15.4 standard is designed to provide reliable data transmission of modest amounts of data up to 100 meters or more while consuming very little power. ZigBee technology takes full advantage of the IEEE 802.15.4 standard and extends the capabilities of this new radio standard by defining a flexible and secure network layer that supports a variety of architectures to provide highly reliable wireless communication. ZigBee technology also offers simplicity and a costeffective approach to building, construction and remodelling with wireless technology. ZigBee is all set to provide the consumers with ultimate flexibility, mobility, and ease of use by building wireless intelligence and capabilities into every day devices. Thus, ZigBee technology is a low data rate, low power consumption, low cost, wireless networking protocol targeted towards automation and remote control applications. The focus of network applications under the IEEE 802.15.4 / ZigBee standard include the features of low power consumption, needed for only two major modes (Tx/Rx or Sleep), high density of nodes per network, low costs and simple implementation. These features are enabled by the following characteristics • • • • • • 2.4GHz and 868/915 MHz dual PHY modes. This represents three license-free bands: 2.4-2.4835 GHz, 868-870 MHz and 902-928 MHz. The number of channels allotted to each frequency band is fixed at 16 channels in the 2.45 GHz band, 10 channels in the 915 MHz band, and 1 channel in the 868 MHz band Maximum data rates allowed for each of these frequency bands are fixed as 250kbps @2.4 GHz, 40 kbps @ 915 MHz, and 20 kbps @868 MHz. Allocated 16 bit short or 64 bit extended addresses. Allocation of guaranteed time slots (GTSs) Carrier sense multiple access with collision avoidance (CSMA-CA) channel access Yields high throughput and low latency for low duty cycle devices like sensors and controls.


• • • • •

Fully “hand-shake” acknowledged protocol for transfer reliability. Low power consumption with battery life ranging from months to years. Energy detection (ED). Link quality indication (LQI). Multiple topologies : star, peer-to-peer, mesh topologies

05. SCOPE:
Increasing network lifetime is a great advantage to keeping the network up even under desolate conditions. We want the network to be robust and resistant to failure. Since the core markets include consumer electronics, energy management and efficiency, health care, home automation, telecommunication services, building automation and industrial automation, an increase in network lifetime would mean increase in efficiency of all the above mentioned and not leaving out the increase in revenue of the company due to a resistant to failure network.

[1] LAN/MAN Standards Committee of the IEEE Computer Society, “IEEE Standard for Information technology— Telecommunications and information exchange between systems— Local and metropolitan area networks— Specific requirements Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs)”, 2006. Holger Karl and Andreas Willig, “Protocols and Architectures for Wireless Sensor Networks”, 2005 John Wiley & Sons, Ltd. ISBN: 0-470-09510-5. Shahin Farahini, “ZigBee Wireless Networks and Transceivers”, 30 September, 2008, NewNes Publishers, ISBN: 0-750-68393-7. András Varga, “OMNeT++ User Guide for v4.2b2”, May 5th, 2011. “INET Framework for OMNeT++ Manual”, 2011. The ZigBee Alliance, <>. OMNeT++ in a Nutshell, <> MiXiM, <>



[4] [5] [6] [7]



APL FFD LQI LR-WPAN MAC Mote Node NWK PAN PHY QOS RFD WLAN WPAN Application Layer Forward Function Device Link Quality Indicator Low Rate – Wireless Personal Area Network Medium Access Layer A ZigBee Coordinator A ZigBee End – Device Network Layer Personal Area Network Physical Layer Quality of Service Reduced Function Device Wireless Local Area Network Wireless Personal Area Network

[1] Andrew Jennings and Daud A. Channa, “Annealing Sensor Networks”, Lecture Notes in Computer Science, Volume 3684, Aug 2005, Pages 581 - 586. Arun S., “Seminar Report on ZigBee”, September 2008. Feng Chen and Falko Dressler, “A Simulation Model of IEEE 802.15.4 in OMNeT++, 2006. Feng Chen, Isabel Dietrich, Reinhard German and Falko Dressler, “An Energy Model for Simulation Studies of Wireless Sensor Networks using OMNeT++”, 2007. Feng Chen, Nan Wang, Reinhard German and Falko Dressler, “Performance Evaluation of IEEE 802.15.4 LR-WPAN for Industrial Applications”, 2008. Sajjad Hussain Shah, Kashif Naseer, Wajid Ali, Sohail Jabbar, Abid Ali Minhas, “Prolonging the Network Lifetime in WSN through Computational Intelligence”, Proceedings of the World Congress on Engineering and Computer Science 2011 Vol I, WCECS 2011, October 19-21, 2011, San Francisco, USA.

[2] [3] [4]



The network lifetime of the ZigBee transceiver is obtained by simulating a ZigBee network on a network simulator. In this project we have opted for OMNeT++. The simulation is first performed first by turning off the sleep cycles and testing the transmissions and the energy consumed. Then the sleep cycles would be introduced and the energy consumed would be recorded. Lastly, the protocol that would be discussed in the theory would be introduced and the energy consumption would be recorded. A graphing utility would be used to plot the results of the energy consumed for every iteration of the test.


The cellular network was a natural extension of the wired telephony network that became persistent during the mid-20th century. As the need for mobility and the cost of laying new wires increased, the motivation for a personal connection independent of location to that network also increased. Coverage of large area is provided through (1-2km) cells that co-operate with their neighbours to create a seamless network. Cellular standards basically aimed at facilitating voice communications throughout a metropolitan area. During themid-1980s, it turned out that an even smaller coverage area is needed for higher user densities and the emergent data traffic. 10.1 Evolution of Low-Rate Wireless Personal Area Network (LR-WPAN) Standardization The IEEE 802.11 working group for Wireless Local Area Network (WLAN) was formed, to create a wireless local area network standard. Whereas IEEE 802.11 was concerned with features such as Ethernet matching speed, long range(100m), complexity to handle seamless roaming, message forwarding, and data throughput of 2-11 Mbps. Wireless personal area networks (WPANs) are used to convey information over relatively short distances. WPANs are focused on a space around a person or object that typically extends up to 10m in all directions. The focus of WPANs is low-cost, low power, short-range and very small size. • • • The high data rate WPAN (IEEE 802.15.3) is suitable for multi-media applications that require very high quality of services. Medium rate WPANs (IEEE 802.15.1/Bluetooth) will handle a variety of tasks ranging from cell phones to PDA communications and have QoS suitable for voice communications. The low rate WPANs (IEEE 802.15.4/LR-WPAN) is intended to serve a set of industrial, residential and medical applications with very low power consumption, with relaxed needs for data rate and QoS. The low data rate enables the LR-WPAN to consume very little power. This feature allows small, power-efficient, inexpensive solutions to be implemented for a wide range of devices.

10.2 Why is it called ZigBee? It has been suggested that the name evokes the haphazard paths that bees follow as they harvest pollen, similar to the way packets would move through a mesh network. Using communication system, whereby the bee dances in a zigzag pattern, worker bee is able to share information such as the location, distance, And direction of a newly discovered food source to her fellow colony members. Instinctively implementing the ZigBee Principle, bees around the world actively sustain productive itchiness and promote future generations of Colony members. 10.3 Device Types ZigBee devices are required to conform to the IEEE 802.15.4-2006 Low-Rate Wireless Personal Area Network (WPAN) standard. ZigBee wireless devices are expected to transmit 10-75 meters, depending on the RF environment and the power output consumption required for a given application, and will operate in the unlicensed Worldwide (2.4GHz global, 915MHz Americas or 868 MHz Europe). The data rate is250kbps at 2.4GHz, 40kbps at 915MHz and 20kbps at 868MHz. There are three different ZigBee device types that operate on these layers in any self-organizing application network. These devices have 64-bit IEEE addresses, with option to enable shorter addresses to reduce packet size, and work in either of two addressing modes – star and peer-to-peer. • The ZigBee (PAN) coordinator node: The most capable device, the coordinator forms the root of the network tree and might bridge to other networks. It is able to store information about the


• •

network. There is one, and only one, ZigBee coordinator in each network to act as the router to other network. It also acts as the repository for security keys. The Full Function Device (FFD): The FFD is an intermediary router transmitting data from other devices. It needs lesser memory than the ZigBee coordinator node, and entails lesser manufacturing costs. It can operate in all topologies and can act as coordinator. The Reduced Function Device (RFD): This device is just capable of talking in the network; it cannot relay data from other devices. Requiring even less memory, (no flash, very little ROM and RAM), an RFD will thus be cheaper than an FFD. This device talks only to a network coordinator and can be implemented very simply in star topology.

An FFD can talk to RFDs or other FFDs, while an RFD can talk only to an FFD. An RFD is intended for applications that are extremely simple, such as a light switch or a passive infrared sensor; they do not have the need to send large amounts of data and may only associate with a single FFD at a time. Consequently, the RFD can be implemented using minimal resources and memory capacity. 10.4 Network Topologies There are 3 basic topologies that exist in ZigBee networks. These can be mingled together to create more topologies. • • • Star Topology Peer-to-Peer Topology Mesh Topology

Figure 1: Star Topology

Figure 2: Peer - to - Peer Topology


Figure 3: Mesh Topology

10.5 Architecture The LR-WPAN architecture is defined in terms of a number of blocks in order to simplify the standard. These blocks are called layers. Each layer is responsible for one part of the standard and offers services to the higher layers. The layout of the blocks is based on the open systems interconnection (OSI) seven-layer model. The interfaces between the layers serve to define the logical links between layers. The LR-WPAN architecture can be implemented either as embedded devices or as devices requiring the support of an external device such as a PC. An LR-WPAN device comprises a PHY, which contains the radio frequency (RF) transceiver along with its low-level control mechanism, and a MAC sub layer that provides access to the physical channel for all types of transfer.

Figure 4: ZigBee Stack


10.6 Network and Application Support layer: The network layer permits growth of network without high power transmitters. This layer can handle huge numbers of nodes. This level in the ZigBee architecture includes • • • The ZigBee Device Object (ZDO) User-Defined Application Profile(s) The Application Support (APS) Sub-layer.

The APS sub-layer's responsibilities include maintenance of tables that enable matching between two devices and communication among them, and also discovery, the aspect that identifies other devices that operate in the operating space of any device. The responsibility of determining the nature of the device (Coordinator / FFD or RFD) in the network, commencing and replying to binding requests and ensuring a secure relationship between devices rests with the ZDO (ZigBee Define Object). The user defined application refers to the end device that conforms to the ZigBee Standard. 10.7 Physical (PHY) layer: The PHY service enables the transmission and reception of PHY protocol data units (PPDU) across the physical radio channel. The features of the IEEE 802.15.4 PHY physical layer are Activation and deactivation of the radio transceiver, energy detection (ED), Link quality indication (LQI), channel selection, clear channel assessment (CCA) and transmitting as well as receiving packets across the physical medium. 10.8 Media access control (MAC) layer: The MAC service enables the transmission and reception of MAC protocol data units (MPDU) across the PHY data service. The features of MAC sub layer are beacon management, channel access, GTS management, frame validation, acknowledged frame delivery, association and disassociation.

1. 2. 3. 4. 5. The simulation should be a large scale implementation, that is, not less than 200 nodes. At least 3 ZigBee Coordinator cluster heads (CH) should be involved. Attest 30 ZigBee Routers should be involved. At least 60 ZigBee End-devices should be involved. The protocol to be tested would have the following specifications: • If the CH is connected to two routers, and one of them is sleeping, the CH would route the frame through the other Router. • If this has been repeated a certain number of times, the CH would wake the other router and let the previous one sleep and save its battery. • If both of them are asleep, the CH would look at the time they have been sleeping and the battery life they have remaining with them and wake the one who has been asleep the longest with more battery life. • If both of them are awake, the CH would determine who has the longest uptime and the least battery remaining of the two, and would put that node to sleep.


1. A network simulator (OMNeT++) is required to perform the said analysis. 2. The version of OMNeT++ should not be less than 4.1b1. 3. The version should be installed on a Windows 7 running PC meeting the minimum system requirements as specified by Windows 7 and OMNeT++ cumulatively. 4. The version of OMNeT++ should have an integrated package of the frameworks a. INET b. MiXiM

1. We have a 3 acre land as a test bed for the network configuration as attached in the email, where: 1 acre = 4046.85642m2 2. There is a ZigBee network present in the whole area (as shown in Figure 5) to monitor the basic elements namely, humidity, wind speed, soil acidity, wildlife etc. 3. Every Cluster head is plugged in, i.e., unlimited uptime. 4. Every other node is on battery life, and as since this is a ZigBee network, every device (whether FFD or RFD) can support more than 2 year of uptime including sleep states. 5. A packet of information travels from the ZigBee device (source) to the Monitoring Center (Sink).

Figure 5: Proposed Scenario


The current applications of the ZigBee transceiver are numerous, which also leads to much possible advancements and enhancements. Most of the applications in various different criterions require only major enhancement: Network Lifetime Prolongation. The longer we can maintain the nodes alive, i.e., prolonging the battery life in the network, the better the application and hence broaden the scope of the ZigBee Transceiver. We would be using two tools to accomplish this: 1. OMNeT++. This is an open source Network Simulator OMNeT++ is a discrete event simulation environment. Its primary application area is the simulation of communication networks, but because of its generic and flexible architecture, is successfully used in other areas like the simulation of complex IT systems, queuing networks or hardware architectures as well. OMNeT++ provides a component architecture for models. Components (modules) are programmed in C++, then assembled into larger components and models using a high-level language (NED). Reusability of models comes for free. OMNeT++ has extensive GUI support, and due to its modular architecture, the simulation kernel (and models) can be embedded easily into your applications. 2. MiXiM Framework MiXiM is an OMNeT++ modelling framework created for mobile and fixed wireless networks (wireless sensor networks, body area networks, ad-hoc networks, vehicular networks, etc.). It offers detailed models of radio wave propagation, interference estimation, radio transceiver power consumption and wireless MAC protocols (e.g. ZigBee).

1. Install OMNeT++ a. Download OMNeT++ from <> b. Extract the latest version of OMNeT++ on the disk. c. Run the bash terminal. WARNING: Do not install OMNeT++ on any directory which has a space in its name, otherwise the bash terminal would register an error. d. Type ./configure and press Enter. e. After the process is over, type make and press Enter. This is a time consuming process. f. After the process is over, type omnetpp to run OMNeT++. g. If the need to log out arises, type exit and press Enter. 2. Install MiXiM framework over OMNeT++ a. Download the latest version of MiXiM <>. b. Open OMNeT++. c. Go to File-> Import. d. Select the .tar.gz file and press import. e. After the import process completes, press CTRL+B to build all configurations and to append MiXiM to OMNeT++. This is a time consuming process. f. Now, all the modules of MiXiM have been integrated on to OMNeT++ for use.


3. Start a new project file of MiXiM by using the Wizard. 4. Make the topology shown in Figure xx . 5. Build the project. NOTE: Add the record-eventlog = true to the omnetpp.ini file. 6. Run the project and simulate up to 100,000 events. 7. Save the simulation and exit. 8. Open the results sub-folder in OMNeT++ file browser and plot the scalar values. 9. Repeat the process from 5 to 8 for: a. No Sleep cycles introduced into the topology. b. Sleep Cycles introduced into the topology. c. The protocols that is proposed in this project, specific to the cluster heads or ZigBee Coordinators.

When the plots start appearing, we can see how the results vary from one another. It may be quite vague at first but, after filtering is applied to the results, the plots would become clear to reveal the results as shown in Graph 1. The parameter record-eventlog = true enables to capture every moment and every magnitude that happens during a simulation. It is also clear to note that unless until it is required, like here, this parameter puts a lot of load on the CPU hence should be used carefully, since it is capturing the most minute of details happening inside the simulated environment. After filtering, the scalar graphs would show how the network life of the whole ZigBee network decreases until one of the node fails. The simulation would end where one node fails and gives an interrupt flag to stop the simulation. Here the total events would be displayed in the Tkenv status bar.

1. The whole simulation should show that the proposed method increases the network lifetime. 2. The Network Lifetime increased should be of a measurable factor. 3. The model drawn out of this proposal should be implementable and should follow the standards specified by IEEE 802.15.4. 4. Since the battery lifespan specified by ZigBee Alliance is minimum 2 years, no node should die before 2 years simulated time.


The following are the results from the simulations successfully completed in OMNeT++:

Graph 1: Results

1. With no sleep cycles in the ZigBee Network, the network dies very soon. 2. As soon as the sleep cycles are introduced, the Network lives up to the ZigBee standard. 3. When the Network is tested on the protocols that are mentioned in this report, the Network life is increased by approximately 15%, which is a measurable and a considerable amount.

As we can clearly see how the results vary from when no sleep cycles are introduced to when the protocol involving the proposed scenario is introduced. This concludes that our method of approach was correct and properly defined.

Recommendations: 1. The scenario should be followed strictly. Any changes would lead to a different scenario and would need to be evaluated. 2. Since this applies to a ZigBee configuration employing ZigBee end-devices, the same scenario can be evaluated with more ZigBee Routers. It might lead to a better Network Lifetime since we are introducing more intelligent devices into the network premises. Technology alternatives can be used which apply the IEEE 802.15.4 stack in a different way as opposed to ZigBee: 1. 2. 3. 4. CEBus LonWorks Insteon Z-Wave

One should note that an alternative would lead to the same amount of paper work as done here.


[1] Fabrizio Granelli, Dzmitry Kliazovich and Nelsom L. S. Da Fonseca, “Performance Limitations of IEEE 802.15.4 Networks and Potential Enhancements”, 2007. Nicky van Frost, “Simulating Queuing Networks with OMNeT++”, January 23rd, 2004. C. Mallanda, A. Suri, V. Kunchakarra, S. S. Iyengar, R. Kannan and A. Durresi, “Simulating Wireless Sensor Networks with OMNeT++”, 2005. E. Egea-López, J. Vales-Alonso, A. S. Martínez-Sala, P. Pavón-Mariño and J. García-Haro, “Simulation Tools for Wireless Sensor Networks”, Summer Simulation Multiconference – SPECTS 2005. Tao Shu and Marwan Krunz, “Coverage-Time Optimization for Clustered Wireless Sensor Networks: A Power-Balancing Approach”, IEEE/ACM Transactions on Networking, Vol. 18, No. 1, February 2010. Deborah Estrin, Ramesh Govindan, John Heidemann and Satish Kumar, “Next Century Challenges: Scalable Coordination in Sensor Networks”, 1999. Pablo Suarez, Carl-Gustav Renmarker, Adam Dunkels and Thiemo Voigt, “Increasing ZigBee Network Lifetime with X-MAC”, 2008. Myung-Gon Park, Kang-Wong Kim and Chan-Gun Lee, “A Holistic Approach for Optimizing Lifetime of IEEE 802.15.4/ZigBee Networks with Deterministic Guarantee of Real-Time Flows”, 2009.

[2] [3]









To appear in KES Invited Session on Evolutionary and Self-Organizing Sensors, Actuators and Processing Hardware. Jennings A. and Channa D., Melbourne, 2005

Annealing Sensor Networks
Andrew Jennings & Daud Channa
Electrical & Computer Engineering School, RMIT University, Melbourne, Australia {,}

Abstract. With a continuing improvement in the capabilities of intelligence per unit of energy, we should reconsider the organisation of sensor networks. We contend that solutions should be model-free, locally based and need to be highly dynamic in nature. Here we propose an approach inspired by simulated annealing. In the context of several application scenarios we explore the potential for adding intelligence to sensor networks.

1. Introduction
Sensor networks [1] are envisioned to become an integral part of our lives. These networks are being applied to provide various tasks such as surveillance and monitoring systems for commercial and military applications. Applications are being developed to gather process and utilize the information from the surrounding environment as required. These requirements have kept challenging researchers in the design of better architectures and protocols for sensor networks. We now have some early deployment of sensor networks, showing that we have successfully established the basic protocols. These follow two main directions: clustering of nodes [2], and synchronised sleep cycle networks with a flatter structure [3]. Now the main challenge is to establish a wide range of application systems. To deploy applications we need methods of coordination that are efficient both in delivering services and conserving energy. The growth and advancements in technologies and the constant reduction in the size and cost of Micro Electro Mechanical Systems (MEMS) have given rise to a whole new dimension of networking which involves sensors and actuators that are quickly deployable and self organizing. They have resulted in a new dimension in network computing, namely pervasive sensing and control. The ongoing rapid advancements, developments, and research in the sensor and actuator networks only leaves one to foresee that they will soon intervene and associate into all living habitats of humans and their surrounding environment. In most scenarios the network must be functional over long periods of time, it is crucial for the operation, management and continued lifespan of the network to control the behaviour/reaction of the sensors and actuators to the different occurrences of events in the network. The sensors in these networks are limited in energy, memory

and computational facilities, while generally the actuators have an ample supply of resources as their mobility can enable them to recharge thus utilizing their resources to the fullest without energy constraint. The deployment and maintenance of the nodes must be cost effective, because it will be unfeasible to configure these large networks of small devices. The sensor nodes along with the actuators must be self organizing and provide a means of programming and managing the network as a whole, rather than administering individual nodes and actuators.

2 Status & scenarios
Although it is not yet clear which applications are viable for sensor networks, we have selected three scenarios to motivate our work. They serve to highlight the issues that we now face. We aim to create solutions for these situations. 2.1 Pedestrian Crossing Guard

Fig. 1. A proposal for an instrumented pedestrian crossing We aim to improve the safety of pedestrian crossings using sensor networks. How might we prevent the running down of pedestrians? One possible solution is to use an array of pressure sensors on the roadway surface to detect pedestrians walking. This would be in addition to proximity sensors and visual surveillance [5], as illustrated in Figure 1. Through combination of sensor readings, we can improve the accuracy of recognition. If we want to make use of the sensor readings then we need confidence that there is a low probability of false positives otherwise car drivers will not accept the system. With a reliable method of pedestrian detection, we can perhaps move to the next level with these systems. If we can reliably detect a car traveling at a speed likely to

result in a collision, then the crossing system can intervene and communicate with the car - potentially it could also override car braking systems. This would bring the car to a halt. We have the possibility of eliminating the possibility of cars contacting pedestrians at crossings. There may also be a role for coordinating with robot teams to enable more active monitoring of pedestrians – here we need development of team behaviour [6] 2.2 Animal Counting Environmental monitoring was one of the earliest motivations for exploring sensor networks. A typical task is the estimation of animal populations. We would like to know how many of a particular type of animal are within a geographic area. In contrast to urban applications, this setting is very demanding in terms of energy management. Note that the estimation of populations is more difficult than simply tracking animals - we need some confidence of the identity of animals. Is the set of readings for a single animal, or two that are within the same area? 2.3 Perimeter Surveillance This is a classical application of sensory technology. We have a perimeter that we wish to patrol, with video and movement surveillance. To augment this, we would like to deploy proximity sensors to improve accuracy. These scenarios can give us a framework to consider sensor network application approaches. They provide challenges and a range of difficulties. All are real applications that may have some prospect of widespread deployment. At this stage of development of the field, it is important to focus on feasible applications to focus the research.

3 Energy and Intelligence
One of the central tenets of sensor networks has been the need to keep nodes simple and careful in their use of energy. We could not, for example, implement the full TCP/IP stack on sensor nodes. This would be a waste of energy, since the nature of the communication is quite different. Progress in battery technology is painful and very slow. But when we consider intelligence per unit of energy, then progress is quite dramatic. So we should be more open to incorporating intelligence in the sensor node, as long as that results in significant energy conservation.

3.1 Local Resolution

Fig. 2. Directed diffusion (network wide) versus local resolution One of the original proposals for sensor network protocols was "directed diffusion" (DD) [4]. It is a robust protocol that can work in very tough environments. Even with extensive network breakages, it will continue to operate. As Figure 2 illustrates, "interests" are propagated to areas of the network, and "gradients" are used to reinforce the successful delivery of packets across the network. Given the constraints placed on directed diffusion, it is an appealing solution. But how might it change if we allowed local processing, rather than propagating results across the broader sensor network? DD assumes that we have to send this outside the sensor network, but with local processing we can avoid the energy consumption of network wide reporting. This creates both a need for local algorithms that can actually resolve sensor data, and also a means of coordination. There are clear energy advantages in local resolution. Similar difficulties arise in the location of mobile sensor nodes. The author in [7] has presented the use of simulated annealing for this problem. Here we are concerned with organisation and messaging for fixed nodes. 3.2 Model Based, or Model Free? If we want to improve monitoring, then perhaps a more accurate model of the context will help? In the case of the pedestrian crossing, we could develop a tracking model. For example, Kalman filtering could aid following an individual through the network. But if we incorporate this model, what will happen in non-modelled cases? How will our model-based approach react when a person falls over, and lies stationary

on the crossing? In the worst case, we might decide that the crossing is clear, and let the cars proceed. Similarly, in surveillance of a perimeter, we might improve accuracy by statistical training to detect people walking across the field. But will we detect somebody crawling across the space? Once we fix a model for the sensing environment, defining the range of possible targets, then the task of constructing the sensor network is reduced to optimisation. There is no need for further intelligence. So the real challenge for sensor networks is how to deal with unusual data. Consider surveillance when a bunch of leaves falls to the ground. Do we raise an alarm or simply log the event for further processing? If we log it, what priority do we give to the event? This is a familiar problem in AI, bringing us to the very familiar challenges of semantics. How do we deal with images that do not have familiar content? How can we go beyond simple statistical pattern matching? These are very difficult, but also very important problems. 3.3 Dynamics Consider the problem of tracking (and identifying) an animal that gives us unusual readings. Perhaps it is of a size that we have not encountered, or it genuinely is a new entry to the region. Clearly this is important, and we would like to track its trajectory. But in order to do this, we need to estimate velocities and alert the relevant part of the sensor network. Once we have lost contact, it will be difficult to sustain the identity. Remember, we are interested in counting animals, so identity is important. Clearly we need application protocols that can deal effectively with highly dynamic situations.

4 Annealing Sensor Networks
The simulated annealing algorithm is a successful method of searching for optimal solutions in complex spaces. Most importantly, it is model free : any problem can be formulated as an annealing process. In analogy with the process of annealing crystals, it has an associated temperature. At high temperature, large parameter changes are possible, but as the process cools only smaller changes are possible.

Fig. 3. Activation cycles (sleep cycles) for a single node

We propose an approach to organising sensor networks that is inspired by simulated annealing. Regions have a temperature, which indicates the intensity of sensing. Figure 3 illustrates the sleep cycle of a single node. At a low temperature, the nodes cycle only at A, but as the temperature increases we also cycle at B,C,D progressively. Since these cycles are divisions of the fundamental cycle (the A cycle), these schedules do not conflict.

Fig. 4. Network temperature heating process In the event that a node encounters an unexpected stimulus, it can cause a local rise in temperature. If several nodes in a region send this message, then a rapid rise in local temperature can take place. We can allow this temperature to spread rapidly in space if we desire, or rapidly decay. In accordance with the physical analogy, local heating cannot spread vast distances without decay. Figure 4 illustrates the process. To effectively make use of annealing sensor networks, we need local resolution of sensor information. For example, in the case of animal tracking, a local decision is needed on the temperature response. Note that an identification is not needed, but only local decision making. We are investigating how to provide this on the typical processors used for sensor nodes. It certainly seems possible to accomplish this computation on the nodes. Of course over time, we can expect the intelligence/energy quotient to continue to grow. Annealing sensor networks (ASN) are a development of directed diffusion networks. There are some important differences. The adaptive sleep cycle provides for rapid response. Local resolution of control is an important difference. Where directed diffusion incorporates routing, ASN's only advise routing.

5 Discussion
We have proposed an approach to model-free local behaviour for sensor networks. Given that we have no local model of expected behaviour, how can we achieve local resolution? Each node can keep a statistical database of patterns it has encountered. When patterns within a statistical tolerance appear, this can trigger the appropriate behaviour. Fully distributed control of sensor networks in this manner raises some important new issues. How do we maintain the currency of statistical data? How can we make changes to behaviour whilst ensuring network stability? It is interesting that classical problems of semantics come to the forefront when we want to further explore sensor networks. Here the resources we have to bring to the problem are limited. We have an unlimited source of data, through lifelong observation of the world through the network sensors. Potentially we can bring vast computation to the task, through recording and processing offline. But we are limited in human intervention. This leads us to explore computationally intensive approaches. Perhaps we should consider the task as “data mining for sensor organisation methods”.

1. Deborah Estrin, Ramesh Govindan, John Heidemann and Satish Kumar “Next Century Challenges: Scalable Coordination in Sensor Networks” In Proceedings of the Fifth Annual International Conference on Mobile Computing and Networks (MobiCOM '99), August 1999, Seattle, Washington. 2. Wendi Heinzelman, Anantha Chandrakasan, and Hari Balakrishnan “Energy-Efficient Communication Protocols for Wireless Microsensor Networks”, Proc. Hawaiian Int'l Conf. on Systems Science, January 2000. 3. Wei Ye and John Heidemann and Deborah Estrin “An Energy-Efficient MAC Protocol for Wireless Sensor Networks” Proceedings 21st International Annual Joint Conference of the IEEE Computer and Communications Societies, 2002 4. Chalermek Intanagonwiwat, Ramesh Govindan and Deborah Estrin “Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks” In Proceedings of the Sixth Annual International Conference on Mobile Computing and Networks (MobiCOM 2000), August 2000, Boston, Massachusetts. 5. Christian Wohler “Autonomous in situ training of classification modules in real-time vision systems and its application to pedestrian recognition” Pattern Recognition Letters Volume 23 , Issue 11 Pages: 1263 - 1270 (September 2002) 6. H. Van Dyke Parunak, Sven. A. Bruekner & James Odell “Swarming pattern detection in Sensor and Robot Networks” Altarium Institute, Working Document 7. Martinson E. and Payton D. “Lattice Formation in Mobile Autonomous Sensor Arrays”, HRL Laboratories, LLC. 2003-2004

Sign up to vote on this title
UsefulNot useful